Andres, all,
* Andres Freund (andres@anarazel.de) wrote:
> On December 21, 2017 10:18:05 PM GMT+01:00, Greg Stark <stark@mit.edu> wrote:
> >On 21 December 2017 at 14:13, Andres Freund <andres@anarazel.de> wrote:
> >> On 2017-12-21 14:05:07 +0000, Greg Stark wrote:
> >>> I wonder why the "out of range" error doesn't print the actual value
> >>> it's trying to cast. That would help the user here...
> >>
> >> We'd have to mark it as non-leakproof in that case.
> >
> >Damn that's annoying.
> >
> >But..... uh, isn't it already leaking that the value is not in
> >99.99999998% of
> >the bigint range?
>
> Most of the relevant operations have more than one operand, or are aggregates. Especially for actually relevant data
ranges.But yes, this is a way to analyze data, we knew that when adding RLS.
Leakproof functions actually were introduced with security barrier
views, which pre-dated RLS and is what RLS is built on top of.
This doesn't change anything wrt this, of course, just figured I'd
clarify for anyone following the thread.
Thanks!
Stephen